Buckets:
| # Make runs/<name>: the same code, venv and data items as the main run, with its own results, logs and markers. | |
| # ./setup_variant.sh minicpm5_2b # then: cd runs/minicpm5_2b && MODEL=openbmb/MiniCPM5-2B ./run_all.sh | |
| set -euo pipefail | |
| cd "$(dirname "$0")" | |
| NAME=$1 | |
| D=runs/$NAME | |
| mkdir -p "$D/data" "$D/markers" | |
| for f in *.py; do ln -sf "../../$f" "$D/$f"; done | |
| ln -sfn ../../.venv "$D/.venv" | |
| ln -sfn ../../../data/raw "$D/data/raw" | |
| cp run_all.sh "$D/run_all.sh" | |
| cp data/train_pool.jsonl data/eval.jsonl data/calib.jsonl data/summary.json "$D/data/" | |
| touch "$D/markers/prep.done" # same items as the main run (prep is seeded; tokenizer only sets prompt_tokens) | |
| echo "ready: $D" | |
Xet Storage Details
- Size:
- 708 Bytes
- Xet hash:
- e782377d26b5461bfe5891ea490469fbc06acc0d51d559c779d7ce02328b33c8
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.